npcpy

Why we included this project

npcpy keeps you from locking into one model vendor. It wraps Ollama, llama.cpp, and LM Studio along with cloud endpoints behind a single Python API, so a prototype that runs on a small local model can later move to a hosted one as your needs change. Beyond chat completions, it has primitives for assembling multi-agent teams and ships a default toolset (shell, file editing, web search) so agents can act on requests rather than just reply. The NPC Context-Agent-Tool data layer shapes how context is stored and governed, leaning on software structure instead of prompt wording, which matters when compliance has to hold up as prompts get edited. For researchers and small engineering groups touching multimodal models, knowledge graphs, and agent workflows, it's a coherent route from a quick experiment to a running system.